I downloaded and run the Deepstream 6.0 image (nvcr.io/nvdeepstream/deepstream6_ea/deepstream:6.0-ea-21.06-devel). I succeeded running the reference application deeepstream-app with many different configurations but when I tried to set yolov3 as primary inference model I couldn’t see any bounding box around the targets. Only object labels appeared in the top-left corner of the image.
I followed the instructions under the official documentation under “DeepStream_6.0_documentation/DeepStream_docs/text/DS_TLT_integration.html#pre-trained-models” for the Yolov3 TLT 2.0 Model architecture:
cd $DS_TOP/samples/configs/tlt_pretrained_models/ then edit deepstream_app_source1_detection_models.txt, let config-file under primary-gie section point to the model config files you want, then run deepstream-app -c deepstream_app_source1_detection_models.txt
I was wondering if the problem is somehow related to the bounding box coordinates format (now in float data type) and the implementation of NvDsInferParseCustomYOLOV3TLT in /opt/nvidia/deepstream/deepstream-6.0/lib/libnvds_infercustomparser.so.
I attached the following files: